3.5 CONCLUSION
The iterative clustering technique developed in the preceding meets the goal that we established for this chapter. It groups the words of Elman's "toy" language using the similarity metric we established in Chapter 2 -- that is, the comparison of contexts that words actually occur. The result is a set of word classes that fit the syntax of the language well (what we would call the "natural" classes of the language). The technique is quite simple. Simple enough, it appears, to be scalable to a more realistic language corpus.
NOTES